In one study,1 researchers investigated whether keyboard activities from habitual smartphone use offer reliable estimates of rest and activity timing when compared with daily self-reports among healthy individuals.
“Rest-activity patterns are important aspects of healthy sleep and may be disturbed in conditions like circadian rhythm disorders, insomnia, insufficient sleep syndrome, and neurological disorders,” authors wrote. Typically, diaries or actigraphy are used to monitor rest-activity patterns over long periods of time.
A total of 51 individuals used a custom smartphone keyboard to passively and objectively measure smartphone use behaviors. Participants also filled out the Consensus Sleep Diary for 1 week. Markers of this method included the time of the last keyboard activity before a nightly absence of keystrokes and the time of the first keyboard activity following this period, researchers explained.
Analyses showed “high correlations between these markers and user-reported onset and offset of resting period (r ranged 0.74-0.80),” while “linear mixed models could estimate onset and offset of resting periods with reasonable accuracy (R2 ranged 0.60-0.66).”
By implementing this method in longitudinal studies, investigators could monitor disturbances to rest-activity patterns without user burden or the use of additional costly devices. The fully unobtrusive method could be “particularly useful in studies amongst clinical populations with sleep-related problems, or in populations for whom disturbances in rest-activity patterns are secondary complaints, such as neurological disorders,” researchers added.
Advancements in Artificial Intelligence
In an additional analysis,2 researchers sought to perform automated sleep staging via deep learning in the Multi-Ethnic Study of Atherosclerosis (MESA) cohort and validate findings against polysomnography (PSG)—the labor-intensive and expensive gold standard for sleep staging.
One feasible alternative to PSG is wrist wearables, which are small in form and have capabilities to continuously monitor metrics. In the current study, researchers’ scheme used actigraphic activity counts and 2 coarse heart rate measures to conduct multiclass sleep staging. Specifically, “our technique uses a combined convolutional neural network coupled and sequence-to-sequence network architecture to appropriate the temporal correlations in sleep toward classification,” authors wrote.
A total of 608 MESA participants were randomly assigned to nonoverlapping training and validation (n = 200) cohorts. Analyses revealed:
The network led to accuracies of 78.66% and 72.46% for 3-class and 4-class sleep staging, respectively.
The 3-stage classifier was especially accurate at measuring non-REM sleep time (mean [SD] time, predicted: 4.98 [1.26] hours vs actual: 5.08 [0.98] hours from PSG)
The 4-stage classifier led to highly accurate estimates of light sleep time (mean [SD] time, predicted: 4.33 [1.20] hours vs actual: 4.46 [1.04] hours from PSG) and deep sleep time (predicted: 0.62 [0.65] hours vs actual: 0.63 [0.59] hours from PSG)
The method was also feasible for sleep staging using Apple Watch–derived measurements. Overall, “This work demonstrates the viability of high-accuracy, automated multi-class sleep staging from actigraphy and coarse heart rate measures that are device-agnostic and therefore well suited for extraction from smartwatches and other consumer wrist wearables,” researchers concluded.
Nonwearable Sleep Tracking
Another alternative to PSG proposed by researchers at the conference is a nonwearable sleep tracking device.3 Although several leading companies have developed such devices that have attracted public interest, “the accuracy of these devices has either been shown to be poor or the validation tests have not been conducted by independent laboratories without potential conflicts of interest,” authors explained.
Researchers tested one of these devices (Beddit, from Apple Inc) under conditions of normal and restricted sleep to see how it compared with PSG and actigraphy. Thirty-five young adults with a mean age of around 19 years were randomly assigned to go to bed at 10:30 PM (normal sleep) or 1:30 AM (restricted sleep).
The majority of participants (77%) were female, and the experiment was performed in a controlled sleep laboratory environment. Lights on occurred at 7 AM for all participants while “sleep was measured by the early version (3.0) or newer version (3.5) of a non-wearable device that uses a sensor strip to measure movement, heart rate, and breathing,” researchers wrote. PSG, wristband actigraphy, and self-reported data were also collected. In addition, each device’s accuracy was tested against PSG for total sleep time (TST), sleep efficiency (SE%), sleep onset latency (SOL), and wake after sleep onset (WASO).
Investigators found:
Although the early version displayed poor reliability (intraclass correlation coefficients [ICCs] < 0.30), the newer version of the nonwearable device yielded excellent reliability with PSG under both normal and restricted sleep conditions.
Agreement was excellent for TST (ICC, 0.96) and SE% (ICC, 0.98), and for the notoriously difficult metrics of SOL (ICC, 0.92) and WASO (ICC, 0.92).
The newer version significantly outperformed clinical-grade actigraphy (ICCs often in the 0.40 to 0.75 range) and self-reported sleep (ICCs often below 0.40).
Authors expressed surprise that the nonwearable device demonstrated greater agreement with PSG than clinical-grade actigraphy. “Though the field has generally been skeptical of commercial non-wearable devices, this independent validation provides optimism that some such devices would be efficacious for research in healthy adults,” they wrote. “Future work is needed to test the validity of this device in older adults and clinical populations.”
References
1. Druijff-van de Woestijne G, McConchie H, de Kort Y, et al. Behavioural biometrics: using smartphone keyboard activity as a proxy for rest-activity patterns. Presented at: Virtual SLEEP 2021; June 10-13, 2021; Virtual. Abstract 248.
2. Chowdhury S, Song T, Saxena R, Purcell S, Dutta J. AI-supported sleep staging from activity and heart rate. Presented at: Virtual SLEEP 2021; June 10-13, 2021; Virtual. Abstract 250.
3. Hsiou D, Gao C, Pruett N, Scullin M. Validation of a non-wearable sleep tracking device in healthy adults under normal and restricted sleep conditions. Presented at: Virtual SLEEP 2021; June 10-13, 2021; Virtual. Abstract 254.
Source: https://www.ajmc.com/view/sleep-2021-tech-roundup-smartphones-artificial-intelligence-and-more
